How does the pre-training objective affect what large language models learn about linguistic properties?

Anonymous

Abstract: Several pre-training objectives, such as masked language modeling (MLM), have been proposed to pre-train language models (e.g. BERT) with the aim of learning better language representations. However, to the best of our knowledge, no previous work so far has investigated how different pre-training objectives affect what BERT learns about linguistics properties. We hypothesize that linguistically motivated objectives (e.g. MLM) should help BERT to acquire better linguistic knowledge compared to using non-linguistically motivated objectives, i.e. hard for humans to guess the association between the input and the label to be predicted. To this end, we pre-train BERT with two linguistically motivated objectives and three non-linguistically motivated ones. We then probe for linguistic characteristics encoded in the representation of the resulting models. We find strong evidence that there is no actual differences in probing performance between the representations learned by the two different types of objectives. These surprising results question the dominant narrative of linguistically informed pre-training.